Aplicação de Modelos de Linguagem Visual na Classificação de Lesões Tumorais: Um Estudo com PubMedCLIP

Abstract

The application of artificial intelligence (AI) in dentistry faces challenges such as data limitations and a lack of standardization in clinical information, hindering progress, particularly in the diagnosis of oral lesions. With oral cancer representing a significant portion of neoplasia cases in Brazil, early detection is essential, and AI can play a crucial role in this process. This study investigated the capacity of the pre-trained visual language model, PubMedCLIP, to classify malignant and benign tumor lesions, achieving AUC-ROC and F1-Score values of 0.9900 and 0.9665, respectively. While the results are promising, the scarcity of clinical data limits the full utilization of the model, which was designed for multimodal tasks.